Structural analysis of concrete shells using deep learning methods

Pollet, Maxime; Shepherd, Paul; Hawkins, Will; Costa, Eduardo

Abstract

The structural behaviour of concrete shells is complex, which typically makes their design and production more difficult than prismatic structures. The Finite Element (FE) method is often used for the structural analysis of shells, but obtaining accurate results can be computationally expensive. The present research investigates the use of deep learning techniques to estimate rapidly and accurately the structural behaviour of concrete shells. While these models require a large initial time investment to generate a training dataset and to fit the models, they can then make predictions in a few seconds. Using a flooring thin-shell system as a test-case, a dataset of 20,000 shells with varying spans, heights, thicknesses, and material properties was generated. Linear FE analysis was used to determine the stresses and the buckling factor of the shells under a load case combining self-weight and live loads. Two types of deep learning models, a Multilayer Perceptron (MLP) and a Convolutional Neural Network (CNN) were trained to predict the stress and the buckling behaviour of shells. The results obtained highlight the ability of deep learning models to predict rapidly and accurately the stresses and the buckling factor of concrete shells, as the errors measured are consistently below 2%.

• Presentation Conference Type: Conference Paper (published)
• Conference Name: International Association for Shell and Spatial Structures
• Start Date: Aug 26, 2024
• Acceptance Date: May 21, 2024
• Deposit Date: Jan 13, 2025
• Peer Reviewed: Peer Reviewed
• Book Title: Proceedings of the IASS 2024 Symposium
• Keywords: Buckling,Concrete,Convolutional Neural Network,Deep Learning,Finite Element Analysis,Machine Learning,Multilayer Perceptron,Shells,Stress,Structural analysis
• Public URL: https://uwe-repository.worktribe.com/output/13612019
• Publisher URL: https://people.bath.ac.uk/ps281/research/publications/zurich_preprint2.pdf